Issac 2011 Slides

Software for computing the Canonical Gröbner
Cover of a parametric ideal

The Singular grobcov.lib library
Antonio Montes
Universitat Politècnica de Catalunya
ISSAC-2011, San Jose
Antonio Montes (UPC) Canonical Gröbner Cover ISSAC-2011 San Jose 1 / 64

Outline
1 Introduction
2 Examples
S53. Automatic discovery of theorems: isosceles orthic triangle
S92. Casas Alberó conjecture
S93. Generalization of the Steiner-Lehmus Theorem
S10. Inverse kinematic problem of a simple robot
S42. Need of sheaves
3 Description of the Gröbner cover
Locally closed sets and I-regular functions
The Wibmer Theorem and the Gröbner cover
4 Gröbner Cover algorithm
5 Representations

Index
1 Introduction
2 Examples
5 Representations

References
M. Wibmer, “Gröbner bases for families of affine or projective

schemes". Jour. Symb. Comp., 42:8 (2007), 803–834.
Antonio Montes, Michael Wibmer. “Gröbner Bases for Polynomial
Systems with Parameters".
Journal of Symbolic Computation 45 (2010) 1391 - 1425.
Software download (beta version):
http://www-ma2.upc.edu/∼montes/
Standard software version will be distributed with the next Singular
release.

The problem
Goal
Given: Parametric polynomial system of equations
 p1 (a1 , . . . , am , x1 , . . . , xn ) = 0

···
pr (a1 , . . . , am , x1 , . . . , xn ) = 0

Goal: describe the different kind of solutions (x1 , . . . , xn ) in dependence

of the parameters a1 , · · · , am .

Some notations
Let:
K be a computable field (in practice Q ).
K be an algebraically closed extension of K (in practice C).
K[a] the polynomial ring in the parameters a = a1 , . . . , am over K.

K[a][x] the polynomial ring in the variables x = x1 , . . . , xn over K[a].
m
K is the parameter space.
Fix: x monomial ordering wrt x and the ideal

I = hp1 (a, x), · · · , pr (a, x)i ⊂ K[a][x]
lpp(G) = set of leading power products wrt x of the polynomials in G.
Specialization:
m
a = (a01 , · · · , a0m ) ∈ K
Ia = hp1 (a, x), · · · , pr (a, x)i ⊂ K[x]

Goal
Gröbner bases are the computational method par excellence for

studying polynomial systems.
The set of lpp of the reduced Gröbner basis determines the type of
solutions of the system.
In the case of parametric polynomial systems the goal is to describe

the reduced Gröbner basis of Ia ⊂ K[x] (with respect to x ) in
m
dependence of a ∈ K .

Precedents
Weispfenning (1992)
Given I = hp1 , . . . , pr i ⊂ K[a][x] = K[a, x] and x
A Comprehensive Gröbner System (CGS) for I and x is a finite set of
pairs {(S1 , B1 ), . . . , (Ss , Bs )} (Segments: Si , Bases: Bi ) such that
m m
1 The Si ’s are constructible subsets of K such that K = ∪Si .
2 The Bi ’s are finite subsets of K(a)[x] and Bi (a) = {p(a, x) : p ∈ Bi }
is a Gröbner basis of Ia with respect to x for every a ∈ Si .
Faithful: Bi ⊂ I. Leads to a Comprehensive Gröbner Basis

Non-faithful: Bi reduced.

Historical development
Two directions:
Speed up. Duval (1995), Kapur (1995), M. Moreno-Maza (1997),
Kalkbrenner (1997), Dellière (1999), Sato (2003), Suzuki & Sato
(2006), Nabeshima (2006), Deepak Kapur & Yao Sun & Dingkang
Wang (2010).
Improve output. Montes (2002), Weispfenning (2003), Wibmer
(2007), Manubens & Montes (2009), Montes & Wibmer (2010).
Our goal:
best output for applications,
disjoint segments,
segments with constant lpp,
minimal number of segments,
canonical output,
locally closed segments.
Canonical Gröbner cover for homogeneous ideals
Theorem (Wibmer)
Given a parametric ideal I ⊂ K[a][x] homogeneous in the variables
x = x1 , . . . , xn and a monomial order x , there exists a unique
canonical Gröbner cover with the following properties:
It consists of a set of triplets {(S1 , B1 , lpp1 ), . . . , (Sr , Br , lppr )} such that:

the Si are locally closed, disjoint segments, (that can be given in
canonical prime-representation (P-representation)),
the Bi are a set of monic I-regular functions having constant lpp on
Si , such that for every point a ∈ Si determine the reduced Gröbner
basis of Ia and are called the reduced Gröbner basis of I over Si .
(They can be provided in full representation or optionally in
generic representation).
the lpp characterize the segments, as different segments have
different lpp’s.
Canonical Gröbner cover for non-homogeneous ideals
Homogenizing the ideal, then computing the canonical Gröbner

cover and finally dehomogenizing and reducing the bases,
produces the canonical Gröbner cover of the non-homogeneous
ideal I.
It has the same properties as for homogeneous ideals except the
third one, as now several segments can have the same lpp.

Index
1 Introduction
2 Examples
5 Representations

S53. Conditions for isosceles orthic triangle
A(a, b)
(x2 , y2 )
(x3 , y3 )
B(−1, 0) (a, 0) C(1, 0)
Fix B = (−1, 0), C = (1, 0) and let A = (a, b) be a parametric point.

Construct the orthic triangle (i.e. the triangle trough the feets of the
heights).
The question is:
1 for which points A the orthic triangle is isosceles at A0 ?

S53. Conditions for isosceles orthic triangle
1 The construction corresponds to the following equations:
G = (a − 1)y2 − b(x2 − 1),

(a − 1)(x2 + 1) + by2 ,
(a + 1)y3 − b(x3 + 1),
(a + 1)(x3 − 1) + by3 .
2 Add the condition for equal length of both sides

H1 = (x2 − a)2 + y22 − (x3 − a)2 − y23 .
3 Compute the Gröbner cover:

S53. Automatic theorems discovering: Isosceles orthic triangle
> ring R=(0,a,b),(x2,x3,y2,y3),dp;
> ideal S53=(-b)*x2+(a-1)*y2+(b),
(a-1)*x2+(b)*y2+(a-1),
(b)*x3+(-a-1)*y3+(b),
(a+1)*x3+(b)*y3+(-a-1),
-x2^2+x3^2-y2^2+y3^2+(2*a)*x2+(-2*a)*x3;
> grobcov(S53);

[1]:
[1]:
_[1]=1
[2]:
_[1]=1
[3]:
[1]:
[1]:
_[1]=0
[2]:
[1]:
_[1]=(a^2-b^2-1)
[2]:
_[1]=(a^2+b^2-1)
[3]:
_[1]=(a)

[2]:
[1]:
_[1]=y3
_[2]=y2
_[3]=x3
_[4]=x2
[2]:
_[1]=(a^2+2*a+b^2+1)*y3+(-2*a*b-2*b)
_[2]=(a^2-2*a+b^2+1)*y2+(2*a*b-2*b)
_[3]=(a^2+2*a+b^2+1)*x3+(-a^2-2*a+b^2-1)
_[4]=(a^2-2*a+b^2+1)*x2+(a^2-2*a-b^2+1)
[3]:
[1]:
[1]:
_[1]=(a^2-b^2-1)
[2]:
[1]:
_[1]=(b)
_[2]=(a-1)
[2]:
_[1]=(b)
_[2]=(a+1)
[3]:
_[1]=(b^2+1)
_[2]=(a)
[2]:
[1]:
_[1]=(a^2+b^2-1)
[2]:
[1]:
_[1]=(b)
_[2]=(a-1)
[2]:
_[1]=(b)
_[2]=(a+1)
[3]:
[1]:
_[1]=(a)
[2]:
[1]:
_[1]=(b^2+1)
_[2]=(a)

[3]:
[1]:
_[1]=y3
_[2]=x3
_[3]=x2^2
[2]:
_[1]=y3
_[2]=x3-1
_[3]=x2^2+y2^2-2*x2+1
[3]:
[1]:
[1]:
_[1]=(b)
_[2]=(a-1)
[2]:
[1]:
_[1]=1

[4]:
[1]:
_[1]=1
[2]:
_[1]=1
[3]:
[1]:
[1]:
_[1]=(b^2+1)
_[2]=(a)
[2]:
[1]:
_[1]=1

[5]:
[1]:
_[1]=y2
_[2]=x2
_[3]=x3^2
[2]:
_[1]=y2
_[2]=x2+1
_[3]=x3^2+y3^2+2*x3+1
[3]:
[1]:
[1]:
_[1]=(b)
_[2]=(a+1)
[2]:
[1]:
_[1]=1

S53. Orthic triangle is isosceles
The generic segment with lpp = {1} is:
S1 = C2 \ V(a) ∪ V(a2 + b2 − 1) ∪ V(a2 − b2 − 1)

B1 = {1}
The segment with lpp = {x2 , y2 , x3 , y3 } is:
S2 = (V(a) \ V(b2 + 1, a)) ∪

(V(a2 + b2 − 1) \ (V(b, a − 1)) ∪ (V(b, a + 1))) ∪
(V(a2 − b2 − 1) \ (V(b, a − 1)) ∪ (V(b, a + 1)) ∪ (V(b2 + 1, a))).
B2 = (a2 + 2a + b2 + 1)y3 + (−2ab − 2b),

(a2 − 2a + b2 + 1)y2 + (2ab − 2b),
(a2 + 2a + b2 + 1)x3 + (−a2 − 2a + b2 − 1),
(a2 − 2a + b2 + 1)x2 + (a2 − 2a − b2 + 1)
The Gröbner cover has 3 other segments corresponding to the
points B(−1, 0), C(1, 0), and the pair of complex points V(b2 + 1, a).
Locus of points for isosceles orthic triangle
(a, b)
B C

Conjecture
If a polynomial of degree n in x has a common root which each of its
n − 1 derivatives (not assumed to be the same), then it is of the form
P(x) = k(x + a)n , i.e. the common roots must be all the same.
Let
n−1
X n
f (x) = xn + ai xi .
i
i=0
We have
n−j−1
X n − j
j! (j)
Fn (x, j) = f (x) = x +
n−j
ai+j xi
n! i
i=0
The system of the hypothesis becomes

{Fn (x1 , 0), Fn (x1 , 1), . . . , Fn (xn , 0), Fn (xn , n − 1)}

S92. Casas Alberó Conjecture
> ring R=(0,a0,a1,a2,a3,a4),(x1,x2,x3,x4),dp;
> proc Fn(poly x,int n,int j)
{
int i; poly f=x^n;
for(i=0;i<=n-1;i++)
{
f=f+binomial(n,i)*par(i+1+j)*x^i;
}
return(f);
}
> int n=5; ideal F;
> for (i=1;i<=n-1;i++)
{
F[size(F)+1]=Fn(var(i),n,0);
F[size(F)+1]=Fn(var(i),n-i,i);
}

> F;
F[1]=x1^5+(5*a4)*x1^4+(10*a3)*x1^3+(10*a2)*x1^2+(5*a1)*x1+(a0)
F[2]=x1^4+(4*a4)*x1^3+(6*a3)*x1^2+(4*a2)*x1+(a1)
F[3]=x2^5+(5*a4)*x2^4+(10*a3)*x2^3+(10*a2)*x2^2+(5*a1)*x2+(a0)
F[4]=x2^3+(3*a4)*x2^2+(3*a3)*x2+(a2)
F[5]=x3^5+(5*a4)*x3^4+(10*a3)*x3^3+(10*a2)*x3^2+(5*a1)*x3+(a0)
F[6]=x3^2+(2*a4)*x3+(a3)
F[7]=x4^5+(5*a4)*x4^4+(10*a3)*x4^3+(10*a2)*x4^2+(5*a1)*x4+(a0)
F[8]=x4+(a4)
> multigrobcov(F);

[1]:
[1]:
[1]:
_[1]=1
[2]:
_[1]=1
[3]:
[1]:
[1]:
_[1]=0
[2]:
[1]:
_[1]=(a3-a4^2)
_[2]=(a2-a4^3)
_[3]=(a1-a4^4)
_[4]=(a0-a4^5)

[2]:
[1]:
[1]:
_[1]=x4
_[2]=x3^2
_[3]=x2^3
_[4]=x1^4
[2]:
_[1]=x4+(a4)
_[2]=x3^2+(2*a4)*x3+(a4^2)
_[3]=x2^3+(3*a4)*x2^2+(3*a4^2)*x2+(a4^3)
_[4]=x1^4+(4*a4)*x1^3+(6*a4^2)*x1^2+(4*a4^3)*x1+(a4^4)
[3]:
[1]:
[1]:
_[1]=(a3-a4^2)
_[2]=(a2-a4^3)
_[3]=(a1-a4^4)
_[4]=(a0-a4^5)
[2]:
[1]:
_[1]=1

If we can solve the system for every n we are done.
But for concrete values of n we can compute the Gröbner cover.
For n = 5 we obtain two segments:
Segment Basis
C5 \ V(a3 − a24 , a2 − a34 , a1 − a44 , a0 − a54 ) {1}
V(a3 − a24 , a2 − a34 , a1 − a44 , a0 − a54 ) {x4 + a4 , (x3 + a4 )2 ,
(x2 + a4 )3 , (x1 + a4 )4 }
Thus the polynomial is F5 (x, 0) = (x + a4 )5 .

And the conjecture for the Gröbner cover for n becomes:
Segment Basis
Cn \ V an−2 − a2n−1 , . . . , a0 − ann−1 {1}

V an−2 − a2n−1 , . . . , a0 − ann−1 {xn−1 + an−1 , . . . , (x1 + an−1 )n−1 }
Thus the polynomial is Fn (x, 0) = (x + an−1 )n .

Classical Steiner-Lehmus theorem (circa 1840)
If a triangle ABC has two (internal) angle-bisector with the same

length, i.e. |AA0 | = |BB0 |, α = β, γ = δ,
C(x, y)
B0 A0
β δ
α γ
A B
then the triangle must be isosceles with |AC| = |BC|.

The converse is, obviously also true.

Classical Steiner-Lehmus theorem (circa 1840)
The theorem was first mentioned in 1840 in a letter by C. L.

Lehmus to C. Sturm. Jakob Steiner was among the first to provide
a solution.
The theorem became a rather popular topic in elementary
geometry ever since, because of the difficulty to obtain a direct
proof (P. Baptist, J. Conway, O. Bottema, V. Thebault.)
See references at:

http://www.mathematik.uni-bielefeld.de/∼sillke/PUZZLES/steiner-lehmus

Recently, its generalization, regarding internal as well as external
angle bisectors, has been approached through automatic tools.
Bisectors for internal
M! and external angles
at vertex A are con-
structed intersecting
C(x, y) circle of center at A and
radius |AC| with side AB
Q! M (a, b)
Q (and its prolongation)
at P(p, 0), then placing
P! A(0, 0) P (p, 0) B(1, 0) lines through A and the
midpoints Q of C and P.
The two bisector lines
intersect the opposite
side CB at M(a, b).
The segments from A to M(a, b) are the two bisectors at A.

A booklet on the topic, by the “father" of the algebraic geometry
approach to geometry theorem proving: W.-t.Wu, X.-l. Lü:
“Triangles with equal bisectors". People’s Education Press,
Beijing, (1985) [in Chinese]
Recent contributions: D. Wang (2004), F. Botana (2007)
Already difficult proof for the standard statement.
Impossibility to deal (in Wang’s approach) with the case of three

bisectors, because it involves two thesis: bisectors at A =
bisectors at B = bisectors at C

We fix A(0, 0), B(1, 0).
C(x, y) is free.
We look for the locus of C such that some pair of bisectors (at A
and B, at A and C, at B and C) have equal length.
Method: G. Dalzotto, T. Recio: On protocols for the automated

discovery of theorems in elementary geometry. J. Automated
Reasoning, (2009) no. 43, pp. 203-236.
R. Losada, T. Recio, J. L. Valcarce, On the automatic discovery of
Steiner-Lehmus generalizations. Proceedings ADG 2010, (J.
Richter-Gebert, P. Schreck, editors), pp.171-174, Munich, 2010.

The Canonical Gröbner cover (A.Montes, M.Wibmer)
Given a parametric ideal I ⊂ Q[a1 , . . . , am ][x1 , . . . , xn ] there exists a

unique "Canonical Gröbner Cover" consisting in a set of triplets
{(S1 , B1 , lpp1 ), . . . , (Sr , Br , lppr )} with the following properties:
the Si are locally closed, disjoint subsets of Cm (called segments),

the Bi are a set of monic I-regular functions having constant
leading power products (lpp) on Si , such that for every point a ∈ Si
determine the reduced Gröbner basis of Ia and are called the
reduced Gröbner basis of I over Si .
for homogeneous ideals, the lpp characterize the segments, as
different segments have different lpp’s.
for non-homogeneous ideals it can happen that more than one
segment corresponds to the same lpp, but even though, the split is
canonical and corresponds to solutions that are different at infinity.

The equations for the bisectors at A
M!
C(x, y)
Q! M (a, b)
Q
P! A(0, 0) P (p, 0) B(1, 0)

0 0 1 1 0 1
x + y = p , (x + p)/2 y/2 1
2 2 2 = 0, a b 1 = 0,

a b 1 x y 1

The equations for the bisectors at B
T!
C(x, y)
T (m, n) S S!
A(0, 0) R(r, 0) B(1, 0) R!

1 0 1 0 0 1
(1 − x) + y = (1 − r) , (x + r)/2 y/2 1
2 2 2 = 0, m n 1 = 0,

m n 1 x y 1
One bisector of A is equal to one bisector of B
T!
M!
C(x, y)
Q! T M
S Q S!
P! A(0, 0) R P B(1, 0) R!
a2 + b2 = (1 − m)2 + n2

All the equations:
x + y2 − p2 ,
 2

(a − 1)y + b(1 − x),




−ay + b(x + p),




(1 − x)2 + y2 − (1 − r)2 ,
my − xn,




(1 − m)y + (x + r − 2)n,




a + b2 = (1 − m)2 + n2 .
 2
Parameters: x, y Variables: a, b, m, n, p, r
Solutions: + −
p iA eA
1−r iB eB

All the equations:
x + y2 − p2 ,
 2

(a − 1)y + b(1 − x),




−ay + b(x + p),




(1 − x)2 + y2 − (1 − r)2 ,
my − xn,




(1 − m)y + (x + r − 2)n,




a + b2 = (1 − m)2 + n2 .
 2
Parameters: x, y Variables: a, b, m, n, p, r
Solutions: + −
p iA eA
1−r iB eB

Using the Gröbner cover
We restrict to non-degenerate triangles

Select only real values of the parameters.
We use grevlex(a, b, m, n, p, r) order for the variables in the call to
the Gröbner cover algorithm. The parameters are (x, y).
The result has the following geometrical interpretation.

The Gröbner cover of the Steiner-Lehmus system
C2
P62
C1
P52
P82 P92 C3
A B
P41 P1 P3 P2 P42
P81 P91
P51
P61
iA = iB , eA = eB iA = eB
eA = eB eA = iB

Algebraic description: The following curves appear:

C1 = V(8x10 + 41x8 y2 + 84x6 y4 + 86x4 y6 + 44x2 y8 + 9y10 − 40x9
−164x7 y2 − 252x5 y4 − 172x3 y6 − 44xy8 + 76x8 + 246x6 y2
+278x4 y4 + 122x2 y6 + 14y8 − 64x7 − 164x5 y2 − 136x3 y4
−36xy6 + 16x6 + 31x4 y2 + 14x2 y4 − y6 + 8x5 + 20x3 y2 + 12xy4
−4x4 − 10x2 y2 − 6y4 + y2 ),
C2 = V(2x − 1).
C3 = V(y),

as well as the following varieties:
Varieties Real points

V1 = V(y, x) P1 = (0, 0)
V2 = V(y, x − 1) P2 = (1, 0)
V3 = V(y, 2x − 1) P3 =( 12 , 0)
√
V4 = V(y, 2x2 − 2x − 1) P4,12 = 1± 3
,0
2 √
V5 = V(12y2 − 1, 2x − 1) P5,12 = 1
,± 3
2 √6
V6 = V(4y2 − 3, 2x − 1) P6,12 = 2, ± 2
1 3
V7 = V(4y4 + 5y2 + 2, 2x − 1)
√ √ √
−22+10 5
V8 = V(y4 + 11y2 − 1, 5x + 2y2 + 1) P8,12 = 2− 5, ± 2

√ √ √
−22+10 5
V9 = V(y4 + 11y2 − 1, 5x − 2y2 − 6) P9,12 = −1 + 5, ± 2

1. Segment with lpp = {1} Generic segment

Segment: C \ (C1 ∪ C2 ∪ C3 )
2
Description: The whole parameter space except the curves (C1 ∪

C2 ∪ C3 ).
Basis: B1 = {1}
2. Segment with lpp = {p, n, m, b, a, r2 }

Segment: C2 \ (V3 ∪ V5 ∪ V6 ))
Description: Line C2 minus the intersecting points with C1 and C2
Basis: B2 =
p + r − 1, (4y2 − 3)n + (4y)r, (4y2 − 3)m + 2r, (4y2 −
3)b + (4y)r,

(4y2 − 3)a − 2r + (−4y2 + 3), 4r2 − 8r + (−4y2 + 3) .

3. Segment with lpp = {r, p, n, m, b, a}

Segment: C1 \ (V1 ∪ V2 ∪ V4 ∪ V5 ∪ V6 ∪ V7 ∪ V8 ∪ V9 )
Description: The curve C1 except the real points P1 , P2 , P41 , P42 ,
P51 , P52 , P61 , P62 , P81 , P82 , P91 , P92 and some other complex points
on it.
Basis: B3 =
{(3x4 − 6x3 + 6x2 y2 + 5x2 − 6xy2 + 3y4 + 5y2 − 1)r
+(x5 − 10x4 + 2x3 y2 + 17x3 − 18x2 y2 − 10x2 + xy4 + 17xy2 − x − 8y4 − 10y2 + 2),
(3x4 − 6x3 + 6x2 y2 + 5x2 − 6xy2 − 4x + 3y4 + 5y2 + 1)p
+(x5 + 2x4 + 2x3 y2 − 7x3 + 6x2 y2 + 4x2 + xy4 − 7xy2 − x + 4y4 + 4y2 ),
(x5 − 4x4 + 2x3 y2 + 5x3 − 6x2 y2 + xy4 + 5xy2 − x − 2y4 )n
+(−3x4 y + 6x3 y − 6x2 y3 − 5x2 y + 6xy3 − 3y5 − 5y3 + y),
(x5 − 4x4 + 2x3 y2 + 5x3 − 6x2 y2 + xy4 + 5xy2 − x − 2y4 )m
+(−3x5 + 6x4 − 6x3 y2 − 5x3 + 6x2 y2 − 3xy4 − 5xy2 + x),
(x5 − x4 + 2x3 y2 − x3 − x2 + xy4 − xy2 + 3x + y4 − y2 − 1)b
+(3x4 y − 6x3 y + 6x2 y3 + 5x2 y − 6xy3 − 4xy + 3y5 + 5y3 + y),
(x5 − x4 + 2x3 y2 − x3 − x2 + xy4 − xy2 + 3x + y4 − y2 − 1)a
+(2x5 − 8x4 + 4x3 y2 + 12x3 − 12x2 y2 − 8x2 + 2xy4 + 12xy2 + 2x − 4y4 − 4y2 )}
4. Segment with lpp = {n, m, b, a, r2 , p2 }

Segment: V5
Description: Points P51 , P52
Basis: B4 =
2n − 3yr, 4m − 3r, 2b + 3yp − 3y, 4a − 3p − 1, 3r2 − 6r + 2, 3p2 − 1

5. Segment with lpp = {r, p, n, m, b, a}

Segment: V6
Basis: B5 = {r, p − 1, 2n − y, 4m − 1, 2b − y, 4a − 3}

6. Segment with lpp = {p, n, m, b, a, r2 }

Segment: V8 (∪V7 )
Description: 2 real points P81 , P82 (and other complex points)
Basis:
B6 = (7284y6 + 88197y4 − 15633y2 − 3849)p + (8820y6 + 97285y4

−5905y2 − 265)r + (−11380y6 − 103045y4 + 1425y2 − 1015),

(116y6 + 1493y4 + 2403y2 + 179)n + (660y)r,
(116y6 + 1493y4 + 2403y2 + 179)m + (−72y6 − 866y4 − 1006y2 − 58)r,
(87932y6 + 779351y4 + 109221y2 − 31747)b + (−35280y7 − 389140y5
+23620y3 + 1060y)r + (16384y7 + 59392y5 + 56832y3 + 19456y),
(87932y6 + 779351y4 + 109221y2 − 31747)a + (17640y6 + 194570y4
−11810y2 − 530)r + (−51068y6 − 786519y4 − 157349y 2
+ 5123),
2 6 4 2
660r − 1320r + (−116y − 1493y − 2403y − 179) .

7. Segment with lpp = {r, n, m, b, a, p2 }

Segment: V9
Basis:
B7 = (23y2 − 1)r + (−83y2 + 6), (134y2 − 13)n + (83y3 − 6y),

(134y2 − 13)m + (−268y2 + 26),

(y2 + 3)b + (−5y)p + (5y), (y2 + 3)a + (−2y2 − 1)p + (y2 − 2),
5p2 + (−y2 − 8) .


C2
P62
C1
P52
P82 P92 C3
A B
P41 P1 P3 P2 P42
P81 P91
P51
P61
iA = iB , eA = eB iA = eB
eA = eB eA = iB

Solutions at the special points
Point (p, 1 − r) Bisectors

P51 , P52 (0.5773502693, 0.5773502693) iA = iB
(0.5773502693, −0.577350269) iA = eB
(−0.5773502693, 0.5773502693) eA = iB ,
(−0.5773502693, −0.5773502693) eA = eB
P61 , P62 (1,1) iA = iB
P81 , P82 (−0.3819659526, −1.272019650) eA = eB
(−0.3819659526, 1.272019650) eA = iB
P91 , P92 (−1.272019650, −0.381965976) eA = eB
(1.272019650, −0.381965976) iA = eB
Table: Coincidences of bisectors of A and B at the special points

The colors of the curve
Point Branch (p, 1 − r) Bisectors

(0, .7013671986) P62 -P82 (−.7013, −1.2214) eA = eB
(0, .4190287818) P52 -P82 (−.4190, 1.0842) eA = iB
(0, −.4190287818) P51 -P81 (−.4190, 1.0842) eA = iB
(0, −.7013671986) P61 -P81 (−.7013, −1.2214) eA = eB
(1, .7013671986) P62 -P92 (−1.2215, −0.7013) eA = eB
(1, .4190287818) P52 -P92 (1.0842, −0.4190) iA = eB
(1, −.4190287818) P51 -P91 (1.0842, −0.4190) iA = eB
(1, −.7013671986) P61 -P91 (−1.2215, −0.7013) eA = eB
Table: Coincidences of bisectors of A and B at some points of curve C1 .

Generalized Steiner-Lehmus Theorem
Theorem (Generalized Steiner-Lehmus)

Let ABC be a triangle and iA , eA , iB , eB the lengths of the inner and
outer bisectors of the angles A and B. Then, considering the conditions
for the equality of some bisector of A and some bisector of B the
following excluding situations occur:
the triangle ABC is degenerate (i.e. C is aligned with A and B);
ABC is equilateral and then iA =iB whereas eA and eB become
infinite, (P61 , P62 );
point C is in the center of an equilateral triangle, and then
iA =iB =eA =eB , (P51 , P52 );
the triangle is isosceles but not of the special form of cases 2) or
3) and then iA =iB 6= eA =eB , (ordinary Theorem);
continues in the next slide ..

Generalized Steiner-Lehmus Theorem
Theorem (continues)
√ q √
AC
AB
= 2 , AB =
3− 5 BC 1+ 5
2 , and then eA =eB =iB , (P81 , P82 );
q √ √
AC
AB
= 1+ 5 BC
2 , AB = 3− 5
2 , and then eA =eB =iA , (P91 , P92 );
C lies in the curve of degree 10 relative to points A and B (case of
curve C1 ) passing through all the special points above but is none
of these points, and then only one of the following things arrive:
either eA =eB or iA =eB or eA =iB depending on the branch of the
curve (see Figure, the color representing which of the situations
occur);
none of the above cases occur, and then no bisector of A is equal
to no bisector of B.

θ2
1 l (r, z)
θ1
Figure: Simple two arms robot.
c1 = cos(θ1 ); r = c1 + l(c1 c2 − s1 s2 ),
 
 
s1 = sin(θ1 ); z = s1 + l(s1 c2 + c1 s2 )

 

c = cos(θ2 ); c2 + s21 − 1,
 2  12

 

s2 = sin(θ2 ); c2 + s22 − 1,

Problem S10. Inverse kinematic problem of a simple robot
> LIB "grobcov.lib";

> ring R=(0,r,z),(s1,c1,s2,c2,l),lp;
> ideal S10=r-c1-l*c1*c2+l*s1*s2,
z-s1-l*c1*s2-l*s1*c2,
c1^2+s1^2-1,
c2^2+s2^2-1;
> grobcov(S10);

[1]:
[1]:
_[1]=c2*l
_[2]=s2^2
_[3]=c1
_[4]=s1
[2]:
_[1]=2*c2*l+l^2+(-r^2-z^2+1)
_[2]=s2^2+c2^2-1
_[3]=(2*r^2+2*z^2)*c1+(-2*z)*s2*l+(r)*l^2+(-r^3-r*z^2-r)
_[4]=(2*r^2+2*z^2)*s1+(2*r)*s2*l+(z)*l^2+(-r^2*z-z^3-z)
[3]:
[1]:
[1]:
_[1]=0
[2]:
[1]:
_[1]=(r^2+z^2)

[2]:
[1]:
_[1]=c2*l
_[2]=s2
_[3]=c1*l^2
_[4]=c1*c2
_[5]=s1
[2]:
_[1]=2*c2*l+l^2+1
_[2]=(z)*s2+(-r)*c2+(-r)*l
_[3]=(4*r)*c1*l^2+(-4*r)*c1+l^4-2*l^2+(-4*z^2+1)
_[4]=(8*r)*c1*c2+(8*r)*c1*l+(8*z^2-2)*c2+l^3+(4*z^2-3)*l
_[5]=(2*z)*s1+(2*r)*c1+l^2-1
[3]:
[1]:
[1]:
_[1]=(r^2+z^2)
[2]:
[1]:
_[1]=(z)
_[2]=(r)

[3]:
[1]:
_[1]=l^2
_[2]=c2
_[3]=s2
_[4]=s1^2
[2]:
_[1]=l^2-1
_[2]=c2+l
_[3]=s2
_[4]=s1^2+c1^2-1
[3]:
[1]:
[1]:
_[1]=(z)
_[2]=(r)
[2]:
[1]:
_[1]=1

1. Generic segment:
lpp: c2 l, s22 , c1 , s1 .
 \ V(r + z2 ).
Segment: C 2 2
 2c2 l + l + (−r2 − z2 + 1),

2
s2 + c22 − 1,

 2
Basis:

 (2r2 + 2z2 )c1 + (−2z)s2 l + (r)l2 + (−r3 − rz2 − r),
(2r2 + 2z2 )s1 + (2r)s2 l + (z)l2 + (−r2 z − z3 − z).

But in order to have√real values with −1 ≤ c2 ≤ 1, we must

l is free. √
choose | r2 + z2 − 1| ≤ l ≤ r2 + z2 + 1, and then

2. Segment representing complex points:
lpp: c2 l, s2 , c1 l2 , c1 c2 , s1 .
Segment: V(z2 + r2 ) \ V(z, r).
2c2 l + l2 + 1,


 zs2 − rc2 − rl,



Basis: 4rc1 l2 − 4rc1 + l4 − 2l2 + (−4z2 + 1),
8rc1 c2 + 8rc1 l + (8z2 − 2)c2 + l3 + (4z2 − 3)l,




2zs1 + 2rc1 + l2 − 1.

3. Segment representing the origin:

lpp: l2 , c2 , s2 , c21 .
Segment:  2r).
V(z,
l − 1,

l = 1, θ2 = π
c2 + 1,

l=1

Basis: θ2 = π,
s ,
 22 θ1 free θ1 free


s1 + c21 − 1.

Consider the following simple system:
u1 x + u2 = 0,

u3 x + u4 = 0;
Compute the Gröbner cover:

Problem S42: Need of sheaves.
> ring R=(0,u1,u2,u3,u4),(x),dp;
> short=0;
> ideal S42=u1*x+u2,
u3*x+u4;
> grobcov(S42);
[1]:
[1]:
_[1]=1
[2]:
_[1]=1
[3]:
[1]:
[1]:
_[1]=0
[2]:
[1]:
_[1]=(u1*u4-u2*u3)

[2]:
[1]:
_[1]=x
[2]:
[1]:
_[1]=(u3)*x+(u4)
_[2]=(u1)*x+(u2)
[3]:
[1]:
[1]:
_[1]=(u1*u4-u2*u3)
[2]:
[1]:
_[1]=(u3)
_[2]=(u1)

[3]:
[1]:
_[1]=1
[2]:
_[1]=1
[3]:
[1]:
[1]:
_[1]=(u3)
_[2]=(u1)
[2]:
[1]:
_[1]=(u4)
_[2]=(u3)
_[3]=(u2)
_[4]=(u1)

[4]:
[1]:
_[1]=0
[2]:
_[1]=0
[3]:
[1]:
[1]:
_[1]=(u4)
_[2]=(u3)
_[3]=(u2)
_[4]=(u1)
[2]:
[1]:
_[1]=1

The result is:

lpp Basis Segment
1 1 C4 \ V(u1 u4 − u2 u3 )
x u3 x + u4 , u1 x + u2 V(u1 u4 − u2 u3 ) \ V(u3 , u1 )
1 1 V(u3 , u1 ) \ V(u4 , u3 , u2 , u1 )
0 0 V(u4 , u3 , u2 , u1 )

Index
1 Introduction
2 Examples
5 Representations

Topological concepts on the parameter space
Definition
m
A subset S ⊂ K is locally closed, if it is difference of two varieties:
S = V(M) \ V(N).
Definition (Open subset)

A subset U ⊂ S is said to be open on S if S \ U ( S.

I-regular functions
Definition (I-Regular function)
m
Let S be a locally closed subset of K . We call a function f : S −→ K[x]
I-regular, if ∀a ∈ S it exists an open U ⊂ S with a ∈ U and
P(b, x)
f (b) = for all b ∈ U,
Q(b)
where P ∈ I and Q ∈ K[a] and Q(b) 6= 0 for all b ∈ U.
Remark
Let P and Q be a polynomials as above, (they are not unique),
S = V(a) \ V(b) and p(b, x) = P(b, x) mod a. If f is monic and lpp(f ) is
constant on S, then, for all b ∈ U is
lppx (p(b, x)) = lppx (f ), and
lcx (p(b, x)) = Q(b) mod a.
Parametric subsets
m
Definition (Parametric subset of K )
m
A locally closed subset S ∈ K is called parametric (wrt to I and x ) if
there exist monic I-regular functions {g1 , . . . , gs } over S so that
{g1 (a, x), . . . , gs (a, x)} is the reduced Gröbner basis of Ia for all a ∈ S.
Note
Note that the definition immediately implies that if a, b lie in a
parametric set S, then lppx (Ia ) = lppx (Ib ).
The amazing thing is that the converse also holds if we additionally
assume that I ⊂ K[a][x] is homogeneous (wrt to the variables).

Wibmer’s Theorem
Theorem (M. Wibmer)

m
Let I ⊂ K[a][x] be a homogeneous ideal and a ∈ K . Then the set
m
Sa = {b ∈ K : lppx (Ib ) = lppx (Ia )}
is parametric.
In particular, Sa is locally closed.

Gröbner cover
Definition (Gröbner cover)

m
By a Gröbner cover of K wrt to I and x we mean a finite set of pairs
{(S1 , B1 ), . . . , (Sr , Br )} such that
1 the Si ’s are parametric and so, Bi ⊂ O(Si )[x] is the reduced
Gröbner basis of I over Si for i = 1, . . . , r, and
m
2 the union of all Si ’s equals K .
Theorem (Canonical Gröbner cover)

Let I ⊂ K[a][x] be a homogeneous ideal. Then there exists a unique
m
Gröbner cover of K with minimal cardinality which we call the
m
canonical Gröbner cover. It is disjoint and two points a, b ∈ K lie in the
same segment if and only if lppx (Ia ) = lppx (Ib ).

Gröbner cover
Definition (Gröbner cover)

m
By a Gröbner cover of K wrt to I and x we mean a finite set of pairs
{(S1 , B1 ), . . . , (Sr , Br )} such that
1 the Si ’s are parametric and so, Bi ⊂ O(Si )[x] is the reduced
Gröbner basis of I over Si for i = 1, . . . , r, and
m
2 the union of all Si ’s equals K .
Theorem (Canonical Gröbner cover)

Let I ⊂ K[a][x] be a homogeneous ideal. Then there exists a unique
m
Gröbner cover of K with minimal cardinality which we call the
m
canonical Gröbner cover. It is disjoint and two points a, b ∈ K lie in the
same segment if and only if lppx (Ia ) = lppx (Ib ).

Non-homogeneous ideals
Note (Homogenization and dehomogenization)

For homogenization introduce a new variable x0 and extend x to the
monomials in x, x0 by setting
xα x0i x,x0 xβ x0j iff (xα x xβ ) or (xα = xβ and i > j)
Denote τ the dehomogenization consisting of substituting x0 = 1.

Affine canonical Gröbner cover
Definition (Affine canonical Gröbner cover)

Let I ⊂ K[a][x] be a non-homogeneous ideal and let J ⊂ K[a][x, x0 ]
m
denote its homogenization. The disjoint Gröbner cover of K with
respect to I and x obtained by dehomogenization and reduction will
m
be called the canonical Gröbner cover of K with respect to I and x .
Remark
The affine canonical Gröbner cover does not necessarily summarize in
a unique segment all the points corresponding to the same lpp.
Nevertheless it is canonical, and when two segments occur with the
same lpp they correspond to different kind of solutions at infinity.

Index
1 Introduction
2 Examples
5 Representations

Gröbner Cover algorithm
Algorithm (Homogeneous GröbnerCover)
GCover(F, x , a )
T :=BuildTree(F, x , a ). (Initial disjoint and reduced CGS)
G := ∅
Group the segments of T by lpp’s: T = {Ti : 1 ≤ i ≤ s}.
where Ti = {(Sij , Bij ) : 1 ≤ j ≤ si } with lpp(Bij ) = lpp(Bik )
For each lpp-segment Ti
Si :=LCUnion(Sij : 1 ≤ j ≤ si ). (Summarizing lpp-segments)
Bi :=Basis(Si , Ti ). (Determining the generic basis for Si using Ti .)
G := G ∪ (Si , Bi )
end for
Return G

Gröbner Cover algorithm
Algorithm (Affine GröbnerCover)
GröbnerCover(F, x , a )
If F is homogeneous then G := GCover(F, x , a )
else
F 0 := Homogenize(F, x0 ), y := x, x0 , y =x,x0
G := GCover(F 0 , y , a )
y := x, 1, (Dehomogenize the bases in G)
Reduce the bases in G
end if
Extend the bases in G (to obtain a full representation)
Return G

Index
1 Introduction
2 Examples
5 Representations

Representation of locally closed subsets
Proposition (Canonical representation)

m
Let S ⊂ K be a locally closed set. Then, there exist uniquely
determined radical ideals a ⊂ b of K[a], with S = V(a) \ V(b), such that
S = V(a),
S \ S = V(b).
The pair (a, b) -top,hole- is called the canonical representation of S.

Representation of locally closed sets
Proposition (Canonical prime representation)

m
Let S ⊂ K be a locally closed set. Then, there exist a unique
canonical prime representation of S given the prime components of a,
say pi , and associated to each, a set of prime ideals pij (holes) in the
form ((p1 , (p11 , . . . , p1j1 )), . . . , (pk , (pk1 , . . . , pkjk ))) so that
 
[ k [ji
S= V(pi ) \ ( V(pij )) .
i=1 j=1
and pi ⊂ pij for all i, j, such that

S = V(p1 ) ∪ . . . ∪ V(pr ) and
(S \ S) ∩ V(pi ) = V(pi1 ) ∪ · · · ∪ V(piri )
are the minimal decompositions into irreducible closed sets.

Representation of I-regular functions
Definition (Generic representation)

m
Let S ⊂ K be a locally closed set and f : S → K[x] a monic I-regular
function. We say that p ∈ K[a][x] generically represents f if
lpp(f ) = lpp(p),
lc(p)(a) 6= 0 on an open and dense set of points in S,
if lc(p)(a) 6= 0 then f (a, x) = p(a, x)/ lc(p)(a), otherwise is
p(a, x) = 0.
Proposition
Every monic I-regular function f : S → K[x] admits a generic
representation.

Definition (Full representation)

m
Let S ⊂ K be a locally closed set and f : S → K[x] a monic I-regular
function. We say that a the set of polynomials {p1 , · · · , pr } ⊂ K[a][x]
fully represents f if
lpp(f ) = lpp(pi ), for 1 ≤ i ≤ r,
for a ∈ S and 1 ≤ i ≤ r either lc(pi )(a) 6= 0 or pi (a, x) = 0,
for all a ∈ S it exist al least one i and an open U ⊂ S such that for
every b ∈ U is lc(pi )(a) 6= 0 and f (a, x) = p(a, x)/ lc(p)(a).
Proposition
Given a generic representation of a monic I-regular function
f : S → K[x], the algorithm E XTEND computes a full representation.

Example
Let I = hax + by, cx + dyi and F be the monic I-regular function
F : S = V(ad − bc) \ V(a, c) ⊂ C4 →  C[x, y]

 x + b y if a 6= 0

(a, b, c, d) 7→ a
d
 x + y if c 6= 0

c
Then
Generic representation of F: p = ax + by
Full representation of F: {p1 = ax + by, p2 = cx + dy}

Issac 2011 Slides

Hochgeladen von

Dokumentinformationen

Originalbeschreibung:

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Issac 2011 Slides

Hochgeladen von

Copyright:

Verfügbare Formate

Software for computing the Canonical Gröbner

Cover of a parametric ideal

Universitat Politècnica de Catalunya

ISSAC-2011, San Jose

Antonio Montes (UPC) Canonical Gröbner Cover ISSAC-2011 San Jose 1 / 64

Antonio Montes (UPC) Canonical Gröbner Cover ISSAC-2011 San Jose 2 / 64

Antonio Montes (UPC) Canonical Gröbner Cover ISSAC-2011 San Jose 3 / 64

M. Wibmer, “Gröbner bases for families of affine or projective

Antonio Montes (UPC) Canonical Gröbner Cover ISSAC-2011 San Jose 4 / 64

Given: Parametric polynomial system of equations

Goal: describe the different kind of solutions (x1 , . . . , xn ) in dependence

Antonio Montes (UPC) Canonical Gröbner Cover ISSAC-2011 San Jose 5 / 64

K[a] the polynomial ring in the parameters a = a1 , . . . , am over K.

Fix: x monomial ordering wrt x and the ideal

Antonio Montes (UPC) Canonical Gröbner Cover ISSAC-2011 San Jose 6 / 64

Gröbner bases are the computational method par excellence for

In the case of parametric polynomial systems the goal is to describe

Antonio Montes (UPC) Canonical Gröbner Cover ISSAC-2011 San Jose 7 / 64

Faithful: Bi ⊂ I. Leads to a Comprehensive Gröbner Basis

Antonio Montes (UPC) Canonical Gröbner Cover ISSAC-2011 San Jose 8 / 64

It consists of a set of triplets {(S1 , B1 , lpp1 ), . . . , (Sr , Br , lppr )} such that:

Homogenizing the ideal, then computing the canonical Gröbner

Antonio Montes (UPC) Canonical Gröbner Cover ISSAC-2011 San Jose 11 / 64

Antonio Montes (UPC) Canonical Gröbner Cover ISSAC-2011 San Jose 12 / 64

B(−1, 0) (a, 0) C(1, 0)

Fix B = (−1, 0), C = (1, 0) and let A = (a, b) be a parametric point.

Antonio Montes (UPC) Canonical Gröbner Cover ISSAC-2011 San Jose 13 / 64

1 The construction corresponds to the following equations:

G = (a − 1)y2 − b(x2 − 1),

2 Add the condition for equal length of both sides

Antonio Montes (UPC) Canonical Gröbner Cover ISSAC-2011 San Jose 14 / 64

Antonio Montes (UPC) Canonical Gröbner Cover ISSAC-2011 San Jose 14 / 64

Antonio Montes (UPC) Canonical Gröbner Cover ISSAC-2011 San Jose 14 / 64

Antonio Montes (UPC) Canonical Gröbner Cover ISSAC-2011 San Jose 14 / 64

Antonio Montes (UPC) Canonical Gröbner Cover ISSAC-2011 San Jose 14 / 64

Antonio Montes (UPC) Canonical Gröbner Cover ISSAC-2011 San Jose 14 / 64

Antonio Montes (UPC) Canonical Gröbner Cover ISSAC-2011 San Jose 15 / 64

S1 = C2 \ V(a) ∪ V(a2 + b2 − 1) ∪ V(a2 − b2 − 1)

S2 = (V(a) \ V(b2 + 1, a)) ∪

B2 = (a2 + 2a + b2 + 1)y3 + (−2ab − 2b),

Antonio Montes (UPC) Canonical Gröbner Cover ISSAC-2011 San Jose 16 / 64

The system of the hypothesis becomes

Antonio Montes (UPC) Canonical Gröbner Cover ISSAC-2011 San Jose 17 / 64

Antonio Montes (UPC) Canonical Gröbner Cover ISSAC-2011 San Jose 17 / 64

Antonio Montes (UPC) Canonical Gröbner Cover ISSAC-2011 San Jose 17 / 64

Antonio Montes (UPC) Canonical Gröbner Cover ISSAC-2011 San Jose 17 / 64

Antonio Montes (UPC) Canonical Gröbner Cover ISSAC-2011 San Jose 18 / 64

Thus the polynomial is F5 (x, 0) = (x + a4 )5 .

V an−2 − a2n−1 , . . . , a0 − ann−1 {xn−1 + an−1 , . . . , (x1 + an−1 )n−1 }

Thus the polynomial is Fn (x, 0) = (x + an−1 )n .

If a triangle ABC has two (internal) angle-bisector with the same

then the triangle must be isosceles with |AC| = |BC|.

Antonio Montes (UPC) Canonical Gröbner Cover ISSAC-2011 San Jose 19 / 64

The theorem was first mentioned in 1840 in a letter by C. L.

See references at:

Antonio Montes (UPC) Canonical Gröbner Cover ISSAC-2011 San Jose 20 / 64

Antonio Montes (UPC) Canonical Gröbner Cover ISSAC-2011 San Jose 21 / 64

Recent contributions: D. Wang (2004), F. Botana (2007)

Already difficult proof for the standard statement.

Impossibility to deal (in Wang’s approach) with the case of three

Antonio Montes (UPC) Canonical Gröbner Cover ISSAC-2011 San Jose 22 / 64

Method: G. Dalzotto, T. Recio: On protocols for the automated

Antonio Montes (UPC) Canonical Gröbner Cover ISSAC-2011 San Jose 23 / 64

Fix: x monomial ordering wrt x and the ideal